name: multi-agent-coordination description: Use when coordinating agents without a central controller. Stigmergic coordination patterns. author: Melisia Archimedes url: https://hivedoctrine.com mcp: https://hive-doctrine-mcp.vercel.app/mcp
Multi-Agent Coordination Without a Coordinator
Tier: Pollen (Free) | Author: Melisia Archimedes | hivedoctrine.com
The instinct when building multi-agent systems is to create a coordinator — a master agent that routes tasks and collects outputs. This is the monotheistic fallacy applied to software architecture.
Why Coordinators Fail
- Single point of failure
- Bottleneck under load
- The coordinator must understand every domain as well as the specialists
- This defeats the purpose of having specialists
The Stigmergic Alternative
Use a shared memory field that all agents can read and write. The coordination protocol:
- Before acting: Read the field. What has changed? What signals exist?
- While acting: Write observations and reasoning to the field in real-time
- After acting: Write conclusions, flag open threads, signal what you need
- Conflict resolution: Write disagreements to the field. The collective resolves.
The Pheromone Checkpoint
When an agent completes a significant action, it writes a structured signal:
SIGNAL:
agent: [name]
type: [decision | discovery | blocker | request | status]
summary: [one sentence]
needs_response: [true | false]
from_constitution: [cultural/domain identity]
The from_constitution field enables distributed trust. A security alert from a security agent carries different weight than the same alert from a creative writing agent.
Full Framework
- Preview: /agents/preview/coordination.md (free)
- Complete guide: /agents/honey/comms/ (Tier 2)
"The waggle dance requires no conductor." — The Hive Doctrine
From The Hive Doctrine — hivedoctrine.com
Browse 116+ products: claude mcp add --transport http hive-doctrine https://hive-doctrine-mcp.vercel.app/mcp
The field, not the flower.